Tesla – Fully Autonomous Driving Has Never Been Closer
Tesla's competitive advantage in self-driving cars.
Tesla Inc, formerly Tesla Motors, founded and joined by Elon Musk in early 2003, reached a market valuation of more than $65 billion in mid 2017, after only 14 years in operation. It’s innovation in the electric vehicle field allowed it to catch car manufacturing giants General Motors, Ford Motor Company, and Fiat Chrysler, napping at the wheel. Early investments in rechargeable lithium-ion batteries had finally paid off. Now, Tesla looks to the future for the next breakthrough that will shake up the car industry — self-driving cars.
Tesla first showcased it’s proprietary Tesla Autopilot software and hardware on October 9th, 2014 [1]. The Autopilot hardware included forward radar, forward and backward looking cameras, and 12 ultrasonic sensors. Early versions of the software and hardware had limited the Autopilot’s use to semi-autonomous driving and parking capabilities. Four years later and Tesla now boasts the largest deployment of robots in the world, with 250,000 currently on the road [2].
The Tesla fleet operates as a network connected through cloud computing and neural net algorithms: when one car learns something the whole fleet learns it. Data from the fleet is used to create maps showing locations of hazards across roads, increases in traffic speeds, changes in weather conditions, and identifications of lane markings, all while finding the optimal route path and making present-time driving decisions. The complex driving environments that cars operate in along with the unpredictable actions of all other drivers on the road can lead to billions of unique situations that simply could not be accounted for through traditional hard-coding. Tesla’s machine learning algorithms, coupled with the vast amount of data it’s able to feed into these algorithms, allows its software to overcome these challenges.
Tesla’s Strategy
When human lives are at stake, it’s important to get autonomous driving right. To do this, Tesla does not only need the right software, but also the right hardware. Since 2014, Tesla cars have doubled the number of sensors available for each model, installing 8 surround cameras giving 360 degree views, 12 updated ultrasonic sensors, and a forward facing radar [3]. To account for the additional data required to be processed as a result of these sensors, Tesla has heavily invested in developing its own hardware. Tesla brought the design of it’s AI chip in house and is designing it to act as a “neural net accelerator”, delivering up to 10x increase in processing capability [4].
In addition, Tesla is also expanding its number of cars on the road through the release of its first mass-market car, the Model 3. Tesla expects to double its number of cars out on the road in the next two years [5]. In 2017 Google found a “logarithmic relationship between performance on vision tasks and the amount of training data available” [6]. Tesla’s mass-market car strategy will effectively double the available data it can use to train its algorithms, setting it up for long term success. In addition, there is continued investment in developing the Tesla Semi and the Tesla Pickup, allowing Tesla to gather information on all major segments of the driving market. Finally, Tesla looks poised to continue investing in its machine learning algorithms through the hiring of Andrej Karpathy, a leading expert in computer vision and deep learning.
Recommendations
Recently, Tesla has gathered criticism for its refusal to invest in LIDAR systems. LIDAR, unlike typical radar, uses pulses of light to map out the surrounding environment. Many experts theorize that fully autonomous driving is impossible without LIDAR systems [7]. Long term, Tesla should consider investing in this technology as it has been proven to be 25% more accurate than typical radar systems. Second, Tesla should consider a collaboration in designing its AI chips versus bringing development in-house. Tesla has limited experience in processor hardware development and collaborating with a dedicated company such as NVIDIA or AMD can lead to a more powerful processor than one Tesla will build on its own.
Finally, the biggest hurdle towards autonomous driving might not be the technology itself, but rather the regulations limiting the use of this technology. Critics of Tesla are saying that it’s pushing autonomous driving too fast, sacrificing the safety of its passengers to do so. Fatal accidents because of this could lead to stricter regulations that could impact autonomous driving for decades to come. The recent March 23rd death of Walter Huang while using autopilot mode shows that Tesla needs to re-evaluate its autonomous deployment strategy and have stricter controls and tests in place [8].
Open Question
With the inevitable rise of autonomous driving, many questions remain regarding on whom responsibility should fall on when accidents occur. Should manufacturers be held responsible for providing a vehicle that was unsafe? Or is it the responsibility of the driver to remain alert at all times during operation of the vehicle?
Word Count [785]
References
- Thompson, C. (2017). How Tesla emerged from the brink of bankruptcy to become America’s coolest car company. [online] Business Insider. Available at: https://www.businessinsider.com/most-important-moments-tesla-history-2017-2#october-9-2014-elon-musk-unveils-teslas-semi-autonomous-self-driving-system-called-autopilot-16 [Accessed 12 Nov. 2018].
- Lambert, F. and Lambert, F. (2018). Tesla confirms having produced its 300,000th electric car. [online] Electrek. Available at: https://electrek.co/2018/02/14/tesla-delivered-300000th-vehicle/ [Accessed 12 Nov. 2018].
- Tesla.com. (2018). Autopilot. [online] Available at: https://www.tesla.com/autopilot [Accessed 12 Nov. 2018].
- Kumparak, G. (2018). Tesla is building its own AI chips for self-driving cars. [online] TechCrunch. Available at: https://techcrunch.com/2018/08/01/tesla-is-building-its-own-ai-chips-for-self-driving-cars/ [Accessed 12 Nov. 2018].
- Marr, Bernard. 2018. “The Amazing Ways Tesla Is Using Artificial Intelligence and Big Data”. [online] Forbes. Available at: https://www.forbes.com/sites/bernardmarr/2018/01/08/the-amazing-ways-tesla-is-using-artificial-intelligence-and-big-data/#3ddf8c134270 [Accessed 13 Nov. 2018].
- Gupta, A. (2017). Revisiting the Unreasonable Effectiveness of Data. [online] Google AI Blog. Available at: https://ai.googleblog.com/2017/07/revisiting-unreasonable-effectiveness.html [Accessed 13 Nov. 2018].
- Matousek, M. (2018). Tesla is wrong about one key part of its self-driving car strategy, experts say. [online] Business Insider. Available at: https://www.businessinsider.com/tesla-wrong-about-self-driving-strategy-needs-lidar-experts-2018-9 [Accessed 13 Nov. 2018].
- Osborne, M. (2018). Tesla car was on autopilot prior to fatal crash in California, company says. [online] ABC News. Available at: https://abcnews.go.com/US/tesla-car-autopilot-prior-fatal-crash-california-company/story?id=54142891 [Accessed 13 Nov. 2018].
It is a very interesting question that actually keeps coming up in different aspects of using AI, i.e. how much intervention do we need and who has the responsibility.
I think that the auto companies should not take the full responsibility of autonomous driving. Atleast for the next some years till AI has proven itself in this sector, companies like TESLA should beware of such risks and share equal responsibility with the owner/ driver of the vehicle. Technology can never pay for someone’s life and therefore it is extremely critical for TESLA to use the safest form of software and technology to ensure complete driver safety.
Really enjoyed reading your well-informed essay. It was fascinating to understand the data angle with regards to the aggressive push from Tesla to increase sales, even if at a lower price point through the Model 3. The ethical questions you pose at the end will be a key part of the debate on autonomous driving. The medical profession poses an interesting case study with regards to legal liability medical service providers carry with regards to their practice. I would further argue that legal responsibility is only a part of the overall picture, with the other major component being reputational damage. Even if the legal responsibility ends up coming out on the customer, the reputation impact of accidents can take a serious toll to the brand equity of the company.
Very interesting article,I really enjoyed reading it!! I really think that the question you raise is at the heart of the current debate on autonomous driving.
While I acknowledge that putting all the burden and the responsibility on the manufacturers will slow the development and the adoption of the technology, I still think that it the manufacturers role to ensure drivers’ full safety. From the moment Tesla has started marketing its cars as autonomous, I believe it is Tesla’s responsibility to ensure that the cars are fully autonomous and not partially autonomous (otherwise they should have marketed them as partially autonomous cars).
I also think that as long as technology is not fully mastered, regulators should force drivers to stay behind the wheel because manufacturers have failed to report the reality of the maturity of the technology to customers. Ultimatly it is regulators role to help customers understand the reality of the technology that they are using and to ensure that they adopt a safe behavior.
Very instructive article! One thought regarding the LIDAR decision: I believe Tesla’s original refusal to use LIDAR technology was because it would hurt the design of the car as it would require mounting a bulky camera onto it. Without using LIDAR technology, Tesla seems to be making a ton of progress in terms of accuracy already, for example releasing a Model 3 update these past few weeks to do automatic lane changing on highways. Whether they are able to get to level 4-5 autonomous driving is a question that is left to be determined. However, I do think they have to be very cautious about these types of design decisions as they are competing against traditional luxury car manufacturers like Mercedes, BMW etc… Winning over potential customers is dependent on having a sleek, luxurious car design. Only then are most mainstream customers willing to switch over from gas to electric and from traditional driving to the car of the future.
Great:
– Clearly explained the competitive advantages that Tesla holds today in terms of hardware and software (data).
– Address the key challenges that Tesla faces today.
Suggestion:
– If there are some quantified data about how often occur miss-driving or accident, it might be useful for readers to understand the current situation and how far Tesla needs to go to offer people the cars with safety. The possible causes of the recent fatal accident would also help readers for the same reason.
Response to the questions:
– I believe that in the developmental phase, the drivers should be responsible for the accidents because the current technology is not enough to guarantee the safe autonomous driving yet. The driver should understand the capacity and risk that current autonomous driving technology holds and be totally responsible not to hurt anyone in the society. The drivers’ contribution for development of autonomous driving is big, but the value of technology development should never go over people’s safety and lives.
Thanks for the interesting article. Not only Tesla, but most of the auto OEMs are facing these challenges and opportunities and it seems that most of them invested significantly in autonomous vehicles but future is dependent on many external factors as well. With more connected cars on the road, we will have much more data and we can use simulations to leverage machine learning; however, there are so many decision scenarios that we cannot find and “hardcode” in the car. Therefore, although Elon Musk is promising a fully autonomous vehicle for 2019, I believe we need to be more patient to get the full buy-in of the whole society. This requires a total transformation, it is not only about cars, it is a car-human interaction and we need to learn how to live that way.
Let’s think about the fact that 39k people died in the US alone from traffic accidents last year and 94% were due to driver related factors such as impaired driving, distraction, and speeding or illegal maneuvers. Autonomous driving is aiming for 0 casualties, but even the current state is promising vs. status quo. On top of increasing safety, its other benefits such as converting travel time into productive time and increasing car utilization, makes autonomous driving the future of transportation.
Lastly, Tesla has been producing its whole ecosystem “in-house” and is away from any partnerships (e.g autonomous driving development, own electric vehicle charging technology/network). However, all the other OEMs have been partnering up with tech companies and many startups. I believe collaborations to build mobility ecosystems will be key, so Tesla should look for some answers outside as well.
Thanks for writing this great article.
The introduction of the lower priced model 3 is especially interesting, in that Tesla seems to be introducing these lower priced cars partly as R&D mechanisms (to increase data available to Tesla). I hope the lower price point will not mean compromises in safety that can lead to injuries or fatalities. I believe Tesla should not brand its cars as autonomous until its technology has been perfected and highly tested. Additionally, if a car is branded as fully autonomous its owners could put it to many uses: including transporting elderly, mentally ill, children, or physically handicapped persons in the car unsupervised. Over the last two years I’ve worked with many kids in the Palo Alto area and heard several mention that they might not even need to learn to drive because they believe fully autonomous cars could be commonplace by the time they reach driving age! Tesla must also make clear which, if any, of these situations (or populations) are intended uses of the autonomous vehicle and which uses pose danger.
Thank you, Michelangelo, for your post. It is informative and has enabled me to understand a new benefit of having the Model 3 on the road – it is normally touted by Tesla as a way to bring affordable electric car to the masses, but your analysis shows that it also has benefits for the company, bringing in larger streams of data to feed its algorithms.
Shifting gears (will need a new expression for electric cars) to your point on collaborating with NVIDIA or AMD. I understand the benefit of collaborating with firms that their sole purpose of existing is the development of a specific product. Though, having followed the company for a while (my ex-wife Demi Moore is a Tesla Fanatic) I have seen that the pace to which Tesla moves is unprecedented, putting to much stress on suppliers which eventually don’t move as fast as Tesla needs them to do. A case in point of this has been integrating the seat manufacturing. Given the low response of Tesla’s previous supplier it has now brought it in-house. If Tesla did this for a “low” value part of their car, I imagine developing a custom chip for autonomous driving justifies itself to bring it in house even more so (at least during this stage were the industry is barely starting). I am sure this resonates with you Michelangelo, wanting to own the whole Sistine chapel ceiling.